In Lesson 8, Jeremy divides the array of learning rates for the different sections of the model by 1000 before passing it to lr_find():
How does lr_find() use these values, and why divide by 1000, which in this case makes the first two values 0?
Hi @immarried,
firstly remember that DL Part 2 2018 uses the old version of the fastai library. Things have changed in the new one.
Note that the first two values are not zero but just small numbers 2e-8 and 2e-7. The reason is that we don’t want to modify the very first layers of the network too much.
As for the other question, take a look at the comments in source here.
